文献类型：期刊文献

中文题名：基于空间学习和情感追踪的多模多目标群搜索算法

英文题名：A Spatial Learning Based SGSO with Emotional Tracking for Multimodal Multi-Objective Optimization

作者：丁亚丹[1];冯翔[1,2];虞慧群[1,2]

机构：[1]华东理工大学信息科学与工程学院,上海200237;[2]上海智慧能源工程技术研究中心,上海200237

年份：2021

卷号：47

期号：6

起止页码：742

中文期刊名：华东理工大学学报（自然科学版）

外文期刊名：Journal of East China University of Science and Technology

收录：Scopus;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：国家自然科学基金(62136003,61772200,61772201);上海市浦江人才计划(17PJ1401900);上海市经信委“信息化发展专项资金”(XXXXFZ-02-20-2463)。

语种：中文

中文关键词：空间学习;情感追踪;群搜索算法;情感因子;多模多目标优化

外文关键词：spatial learning;emotion tracking;group search algorithm;emotion factor;multimodal multi-objective optimization

摘要：为了解决多模态多目标优化问题,寻找与帕累托最优解等效的所有解,通过在基本的群搜索算法中引入社会行为,提出了一种新颖的基于空间学习机制和情感追踪行为的社会群搜索优化算法(MMO;TSGSO)。首先,建立空间学习机制,根据学习到的个体自身位置与最佳个体位置的实时信息,对种群分布状态(离散态、聚合态)进行决策。当种群处于离散态时,采用追随和游走的方式增强算法空间探索能力;随着优化过程的进行,个体彼此影响交互,空间距离逐渐减小,此时种群逐渐聚合,采用动态步长的搜索策略更新个体位置,能实时勘探最优解周围的解,加快算法的收敛速度。其次,引入了情感因子,使一定的个体沿其偏好方向进行情感追踪移动行为,防止算法陷入停滞状态,提高算法求解精度;采用特殊的拥挤距离计算方式和引导进化策略保证算法在决策空间和目标空间的双重多样性。最后,从理论上证明了该算法的收敛性。使用15个多模态多目标优化测试基准函数验证算法的性能,并将其与现有的几个多模多目标优化算法进行性能对比,实验结果验证了本文算法能够有效求解多模多目标优化问题。
In order to solve the multimodal multi-objective optimization problem and find all solutions equivalent to the Pareto optimal solution,this paper proposes a novel group search optimization algorithm(MMO_LTSGSO)based on spatial learning mechanism and emotion tracking behavior by introducing social behavior into the basic group search algorithm.Firstly,a spatial learning mechanism is established and the decision of the population distribution state(discrete state and concentrated state)is made according to the real-time information of the learned individual's own position and the best individual position.When the population is in a discrete state,the following and wandering way is adopted to enhance the space exploration ability of the algorithm.With the optimization process,individuals interact with each other,and the spatial distance gradually decreases.At this time,the population gradually aggregates,and the dynamic step search strategy is used to update the individual position,which can explore the solution around the optimal solution in real time and accelerate the convergence speed of the algorithm.Secondly,in order to prevent the algorithm from falling into stagnation and improve the accuracy of the algorithm,the emotion factor is introduced to make certain individuals track their moving behavior along their preferred direction.Then,special congestion distance calculation and guided evolution strategy are used to ensure the diversity of the algorithm in decision space and target space.Finally,the convergence of the algorithm is proved theoretically,and its performance is verified via 15 multimodal multi-objective optimization test benchmark functions,and is also compared with several existing multimodal multi-objective optimization algorithms.It is shown via the experiments results that the proposed algorithm can effectively solve multimodal multi-objective optimization problems.